← Home

A World Cup of Prediction Markets

Which nations actually built the machinery of betting, pricing, and forecasting?

The final

Australia built the machine. George Julius turned the parimutuel into a thundering acre of shafts, cables and epicyclic gears, the automatic totalisator, that tallied thousands of bets a second and drove the running totals up mechanical thermometer-style indicators. A room-sized real-time computer (1913), decades ahead of the electronic kind. America built the theory, from Bell Labs to the blockchain. It took the field late; a couple of centuries of puritanical disapproval of betting kept the whole side in the tunnel, and once it came out it never stopped scoring.

The road there (knockout recaps)

Quarter-finals

🇦🇺 Australia 3–1 🇯🇵 Japan. Julius opens with the totalisator, Hyndman adds a forecasting goal, and the Sydney Futures Exchange converts the first financial-futures contract launched outside the US (1979). Japan reply through the Dōjima Rice Exchange (Osaka, sanctioned 1730, the first organised futures market), and run the hosts close with Amari's information geometry and Kiyosi Itô's stochastic calculus, the very maths the rest of the tournament is played on.

🇳🇱 Netherlands 2–1 🇩🇪 Germany. The Dutch press with the VOC and the Amsterdam bourse (1602, the first modern stock exchange) and the descending-price flower clock; Germany answers through Gneiting's strictly proper scoring rules, Selten's game theory, and Gnosis (Köppelmann & Stefan George), whose constant-product design seeded the on-chain AMM, but the founders of the share market hold on.

🇺🇸 USA 3–1 🇮🇹 Italy. A game of two halves. De Finetti, the real father of risk-neutral probability (belief priced as coherent, no-arbitrage betting odds), puts Italy ahead early, with Cardano (the first treatise on games of chance) and the Genoese lotto in support. Then the Americans run riot: Kelly's growth-optimal betting, Vickrey's second-price auction, the Iowa Electronic Markets, Hanson's LMSR and Vernon Smith's experimental double auction. They turn it around.

🇫🇷 France 2–1 🇬🇧 UK. Oller's pari-mutuel (he also founded the Moulin Rouge) and Bachelier's 1900 random walk do the scoring, but France's pedigree runs deeper still: Pascal and Fermat invented expected value settling a 1654 gambling dispute, and Laplace marshals the probability behind them. They edge a stacked British side of Bayes, Ramsey, Keynes's beauty contest, Galton's vox populi, and Good & Turing at Bletchley, and a century on Black & Wray's Betfair, the peer-to-peer betting exchange (2000). The result flatters nobody; the UK out-pass everyone and still go out.

Semi-finals

🇦🇺 Australia 2–0 🇳🇱 Netherlands. A Quiggin screamer opens it: rank-dependent (anticipated) expected utility, the theory of how people overweight long-shot probabilities, a direct hit on the favourite–longshot bias the Dutch never closed down. Rob Hyndman's forecasting seals the win. Australia's theoretical goal of the tournament, against the founders of the share market.

🇺🇸 USA 3–2 🇫🇷 France. A classic. Savage, Arrow, and Fama's efficient markets for the US; Walras's imaginary auctioneer and Bachelier for France. The American bench (Pennock, Chen, Wortman Vaughan) proves decisive late on, in what was surely the tougher side of the draw.

The finalists' line-ups

🇺🇸 USA: the bench goes on forever

* called up despite being born abroad (see VAR).

🇦🇺 Australia: yes, a theory line

* born abroad or naturalised.

One Australian tradition never made the team sheet: two-up, the people's parimutuel, a bet on the spin of two coins with no inventor on record (you cannot patent a coin toss). It descends from the British "cross and pile" / "pitch and toss"; the convicts' fondness for it was logged by New South Wales's first judge advocate in 1798, and it spread across the goldfields by the 1850s. The purest binary market in the tournament, and, fittingly, legal nationwide about one day a year.

Group stage (the nations who shaped the game and went home early)

• 🇦🇹 Austria. Friedrich Hayek, whose 1945 "The Use of Knowledge in Society" argues that a market is an information-processing machine, its prices aggregating dispersed knowledge no central planner could gather: the intellectual charter behind every prediction market on this site. Harsh on the federation that wrote the constitution to go out in the group stage. Hayek saw out his career playing for Britain at the LSE.
• 🇮🇱 Israel. Kahneman & Tversky (prospect theory: how people really weight probabilities), Aumann (correlated equilibrium), and Bancor (the first deployed on-chain AMM, 2017). Out on goal difference, and widely held to have been robbed.
• 🇭🇺 Hungary. Von Neumann (the minimax theorem, game theory), Harsanyi (games of incomplete information), Wald (sequential analysis). A golden generation that mostly suited up for other nations.
• 🇷🇺 Russia. Kolmogorov (the 1933 axioms of probability), Markov (Markov chains), Kantorovich (linear programming / optimal allocation): they wrote the rulebook of chance and never turned up to bet, until Michael Egorov's Curve brought the StableSwap AMM on-chain.
• 🇨🇭 Switzerland. The Bernoullis of Basel (the law of large numbers; expected utility and the St Petersburg paradox), the Lausanne School stadium (where Walras and Pareto played), and the Ethereum Foundation in Zug. Neutral hosts who happen to own the ground.
• 🇮🇪 Ireland. Adrian Raftery (Bayesian model averaging; co-author of the canonical proper-scoring-rules paper) and Intrade, Ron Bernstein and John Delaney's election exchange (to 2013); Delaney died on Everest in 2011.
• 🇪🇸 Spain. El Gordo, the Lotería de Navidad (first drawn 1812; the largest lottery in the world by total prize payout), and a formal protest lodged over France's Catalan-born striker.
• 🇬🇷 Greece. Spyros Makridakis and the M-competitions, the long-running forecasting tournaments that benchmark methods head-to-head (and quietly humble the fancy ones); the M5 ran on Australia's Kaggle.
• 🇳🇿 New Zealand. Host of the very first automatic tote (Ellerslie, Auckland, 1913). The assist of the century.
• 🇨🇦 Canada. Keeps having its players poached: Vickrey (Victoria, BC), Tetlock (Toronto), and Myron Scholes (Timmins, Ontario, of Black–Scholes–Merton) all turned out for the USA.
• 🇨🇳 China. Hong Kong's racecourse pari-mutuel pools (the Jockey Club) are among the deepest in the world, so deep that the favourite–longshot bias all but disappears: Busche & Hall found bettors there pricing as if risk-neutral, the very anomaly the rest of the field keeps tripping over. Also keno and the "white-pigeon ticket", though the Great-Wall-lottery tale is folklore and that goal is disallowed.

Several of these squads also carry a lottery, and pooled-stakes gambling is the parimutuel's own ancestor: Augustus drew prizes by lot at Roman feasts; Genoa abstracted betting on the names of elected officials into the 5-of-90 lotto (early 17th century); Elizabeth I held England's first state lottery (drawn 1569); the Dutch Staatsloterij (1726) is often called the oldest still running; Spain's El Gordo (1812) is the biggest by total payout; and a 1612 lottery helped bail out the Jamestown colony. Different mechanism, same idea as this whole site: put the stakes in a pool and let a rule do the rest.

Honourable mentions

A deep tournament leaves talent in the stands:

• Play-money & social markets: the Foresight Exchange / Idea Futures (Sean Morgan, Ken Kittlitz and colleagues, 1994); NewsFutures, later Hypermind (Émile Servan-Schreiber & Maurice Balick); Manifold (Austin Chen and the Grugett brothers); Metaculus (Anthony Aguirre).
• Regulated & retail exchanges: Kalshi (Tarek Mansour & Luana Lopes Lara), the first CFTC-regulated event exchange; PredictIt (Aristotle's John Aristotle Phillips & David Mason, run via Victoria University of Wellington).
• Combinatorial scaling: David Porter (with Ledyard at Net Exchange and on the Policy Analysis Market), and the tractable combinatorial-market-maker work of Dudík, Lahaie & Pennock.
• Order-book microstructure: the statistical-mechanics models of the continuous double auction (Eric Smith and colleagues), built on Gode & Sunder's zero-intelligence traders.
• DeFi liquidity: Alan Lu, who first wrote down x·y=k at Gnosis, and Fernando Martinelli, whose Balancer generalised it to weighted multi-asset pools.
• Calibration: A. Philip Dawid, whose prequential analysis ties truthful sequential forecasting to the scoring rules throughout this site.

VAR is reviewing (contested citizenships & disallowed goals)

• Australia's captain holds three passports. George Julius was born in Norwich, England 🇬🇧; raised and engineering-schooled in New Zealand (Canterbury College); and built the totalisator in Sydney, which he first installed back in Auckland. England claims the birth and New Zealand the upbringing and the debut, but the ruling on the field stands: with the career, the knighthood, and the company all in Sydney, Julius plays for Australia.
• William Vickrey: born in Victoria, British Columbia 🇨🇦; naturalised American; the only economics Nobel laureate born in BC. Played his whole career for 🇺🇸.
• Philip Tetlock (Toronto 🇨🇦) and Yiling Chen (China 🇨🇳) also line up for the USA; Justin Wolfers was born in Papua New Guinea and raised in Sydney, playing for Australia.
• Joseph Oller was Catalan (born in Terrassa 🇪🇸); de Finetti was born in Innsbruck; Daniel Bernoulli in Groningen; Wald in Cluj. Birthplace and shirt rarely agree.
• Vitalik Buterin's AMM goal is disallowed. He suggested and popularised running on-chain exchanges like prediction markets, but invented no market mechanism and did not originate x·y=k. The credit reads Hanson (LMSR) → Gnosis (Köppelmann, constant-product origination) → Uniswap (Adams), with Bancor first to deploy.
• The Netherlands' “Dutch book” goal is disallowed. Despite the name, the sure-loss Dutch book is no Dutch invention: the coherence argument is Ramsey's and de Finetti's, and the “Dutch” is the same disparaging slang as Dutch courage and Dutch treat: bookmakers' jargon of uncertain origin, not a goal for the Oranje.
• The perpetual swap was scored from offshore (BitMEX). The Policy Analysis Market was a market on Middle-East macro indicators, not (as widely misreported) a market on assassinations; we score the design, not the headline.

The mathematics and the sources are on the catalog and in the bibliography (Rhode & Strumpf on the old election markets; Conlon on the totalisator).